I Built an Entire Operating System That Runs in Your Browser — And Its AI Can Rewrite Its Own Code. NexusOS is a fully autonomous AI operating system with 49 apps, a self-healing kernel, and an AI engine that can build new applications from natural language. 100% open source.
What if your OS could think?
Not "think" like a chatbot that generates text.
Think like an entity that can:
See every file on your system
Open, close, and rearrange your applications
Build an entirely new app when you describe what you need
Detect when something crashes... and rewrite its own code to fix it
I've been building this for the past few months. It's called NexusOS, and it runs entirely in your browser.
Yes, that's a full desktop environment. In a browser tab.
Why I Built This
Every "AI-powered" tool I've used follows the same tired pattern:
App exists
Chatbot gets bolted on
Marketing calls it "AI-native"
But the AI can't do anything. It can answer questions. It can't open your file explorer. It can't build a calculator when you need one. It can't notice that your terminal crashed and fix the bug.
I wanted an OS where the AI is the kernel — not a feature.
The Architecture (for the nerds 🤓)
NexusOS is built with:
React 19 + TypeScript — Full component architecture
Vite — Sub-second hot reload
Zustand — State management for the entire OS state (windows, filesystem, processes)
Wllama/GGUF — Local AI inference. No cloud. No API keys required.
Virtual File System — POSIX-like VFS with permissions, symlinks, and undo/redo
The DAEMON Engine
At the heart of NexusOS lives DAEMON — an autonomous intelligence engine woven directly into the kernel.
DAEMON doesn't just respond to prompts. It has a set of 21 OS-level actions it can execute:
OS::OPEN_APP:calculator → Opens the calculator
OS::WRITE_FILE:/home/readme.md → Creates a file
OS::BUILD_APP:{...} → Builds an ENTIRE new React app
OS::SET_WALLPAPER:nebula → Changes the wallpaper
OS::NOTIFY:Build complete → Sends a system notification
When you talk to DAEMON, it doesn't just chat. It acts.
Token-Efficient Context Engine
One of the hardest problems was making AI work with small local models (1-3B parameters). Our solution: a 3-tier adaptive manifest that dynamically adjusts context injection based on query complexity:
Tier When Tokens
Minimal "hello, how are you?" ~80
Standard "open the settings" ~300
Full "build me a todo app" ~500
Compare that to the 3,500+ tokens most AI tools inject per request. We achieved a 75-98% reduction.
This means NexusOS runs smoothly on a Llama 3.2 1B model. On your laptop. No GPU required.
49 Built-In Applications
This isn't a proof of concept with 3 demo apps. NexusOS ships with 49 fully functional applications:
Start Menu
Here's a partial list:
Category Apps
AI & Dev HyperIDE, Neural Forge, DAEMON Chat, Terminal, Model Manager
System Settings, File Explorer, Task Manager, Device Manager, Dashboard
Productivity Notepad, Rich Editor, Calendar, Kanban Board, Pomodoro, Snippets
Media Music Player, Image Viewer, Video Player, Paint, Wallpaper Generator
Utilities Calculator, Weather, Contacts, Password Vault, RSS Reader, Archiver
Every app is a full React component with its own state, window frame, and keyboard shortcuts.
Multi-Provider AI Gateway
NexusOS doesn't lock you into one AI provider.
AI Providers, Connect any AI backend:
✅ OpenAI (GPT-4, GPT-4o)
✅ Anthropic (Claude 3.5)
✅ Google (Gemini)
✅ Groq (blazing fast)
✅ Mistral
✅ OpenRouter
✅ LM Studio (local)
✅ Ollama (local)
✅ GGUF models via Wllama (fully offline)
Or run it with zero API keys. The built-in Wllama engine loads HuggingFace GGUF models directly in WebAssembly.
HyperIDE: A Full Code Editor with AI Copilot
This one's my favorite.
HyperIDE is a complete code editor built into the OS with:
📁 File tree explorer (connected to the VFS)
🎨 Syntax highlighting
💾 Auto-save
🤖 DAEMON Copilot — Click "Explain", "Fix Bugs", or "Refactor" and the AI works on your code
⚡ Integrated terminal
The AI copilot isn't a chatbot sidebar. It's connected to the kernel. When DAEMON writes code, it can save it directly to the filesystem, register it as a new app, and launch it — all in one flow.
Window Management That Actually Works
Multi-Window
Drag, resize, minimize, maximize, close
Window snapping and stacking
Always-on-top pin
Opacity control (glassmorphism)
3 virtual workspaces
Taskbar with live app indicators
Right-click context menus everywhere
The Self-Healing Part
This is where it gets wild.
When DAEMON detects an error in its own output, it runs an ErrorGuard pass:
typescript
const { output, fixApplied, errors } = await errorGuard.guard(
rawResponse, contextualPrompt, fullSystemPrompt, mode
);
ErrorGuard checks for:
Malformed JSON in OS actions
Code syntax errors in generated apps
Missing imports
Hallucinated function calls
If it finds issues, it automatically rewrites the response before it reaches the UI. The user never sees the bug.
Try It Yourself
30-Second Setup
bash
git clone https://github.com/AFKmoney/nexusOS.git
cd nexusOS
npm install
npm run dev
Open http://localhost:3000. That's it. No backend. No Docker. No config files.
With AI (Optional)
Open Settings → AI Providers
Add your OpenAI/Anthropic/Google key
Or: Open Model Manager → Download a GGUF model from HuggingFace
Start talking to DAEMON
What's Next
I'm building in public. Here's the roadmap:
DAEMON Sandbox — Isolated execution environment for AI-generated code
Plugin Marketplace — Community-built apps installable from the App Store
Collaborative Workspaces — Multiple users in the same NexusOS instance
Mobile PWA — Full NexusOS on your phone
Voice Control — "DAEMON, open the terminal and run the build"
This is Open Source
MIT License. Star it, fork it, break it, rebuild it.
⭐ github.com/AFKmoney/nexusOS
(It's still under development, there's is some bugs and unfinished functions, I'm working actively to resolve all the issues)
If you've ever dreamed of building your own operating system — or if you just want to see what happens when AI stops being a chatbot and starts being a kernel — give NexusOS 5 minutes.
I promise you'll want to give it more.
Built by @afkmoney. If this resonated, drop a ⭐ on the repo — it genuinely help